6 ms·

Software specialists at youtube could not say _exactly_ why video-a was sent to users instead of video-b. "Its an algorithm and machine learning." Something w

Software specialists at youtube could not say _exactly_ why video-a was sent to users instead of video-b. "Its an algorithm and machine learning." Something we can't 100% fully explain is already detecting what to send us for entertainment, politics, and information. Right or left, it doesn't matter, both sides have claimed the algorithm has 'helped X' or 'hurt X' This article implies the takeover will happen in 15-30 years. How cute. Why do you think it isnt already here?

safety1st 4y ago
Back in the real world, I think this is the Achilles heel of machine learning. A company has invented a tool which has non-deterministic outputs. Said company is also liable for the outputs of said tool. Inevitably machine learning is going to break laws (already happening) and then the owner of the ML instance is going to get sued for damages and have to think twice about using ML. One of the options will be to just turn off the ML, phase it out etc. which everyone still has the ability to do and that's one of the reasons we're not in Terminator world and probably won't be soon.
mrcartmeneses 4y ago
Because the algorithms are currently “just” transforms based on probability. There’s no understanding present in current ML models.
- OrangeMonkey 4y ago
  Question - how do you think consciousness develops? See, I don't know. So, giving an algorithm the ability to store facts, determine weights, and make decisions that they can see outcomes of... who is to say thats not how consciousness spontaneously develops. Smarter men than me perhaps. But, again, smarter men than me also think we should nuke the internet :)
- IceMetalPunk 4y ago
  Define "understanding" in such a way that predicting what comes next based on experience of how things have happened in the past is not included. In other words, define "understanding" in such a way that humans have it, but these ML models (especially LLMs) don't.

safety1st 4y ago

Back in the real world, I think this is the Achilles heel of machine learning. A company has invented a tool which has non-deterministic outputs. Said company is also liable for the outputs of said tool. Inevitably machine learning is going to break laws (already happening) and then the owner of the ML instance is going to get sued for damages and have to think twice about using ML. One of the options will be to just turn off the ML, phase it out etc. which everyone still has the ability to do and that's one of the reasons we're not in Terminator world and probably won't be soon.

mrcartmeneses 4y ago

Because the algorithms are currently “just” transforms based on probability. There’s no understanding present in current ML models.

OrangeMonkey 4y ago
Question - how do you think consciousness develops? See, I don't know. So, giving an algorithm the ability to store facts, determine weights, and make decisions that they can see outcomes of... who is to say thats not how consciousness spontaneously develops. Smarter men than me perhaps. But, again, smarter men than me also think we should nuke the internet :)
IceMetalPunk 4y ago
Define "understanding" in such a way that predicting what comes next based on experience of how things have happened in the past is not included. In other words, define "understanding" in such a way that humans have it, but these ML models (especially LLMs) don't.